How does ezbot predict the best user experience for every user? We take our checkpoints from our training and validation process to serve predictions at scale. It takes a little learning time before we rapidly begin improving conversion rates. #AI #ecommerce #marketing #startup #conversionrate

Now that checkpoint does something interesting. It doesn't just predict the best possible user experience. It predicts the value of each user experience, sort of like a probability density. If we always select the best user experience, we may never learn something new. We may never realize that for this user there's a better option, or that this experience that people haven't seen yet is really going to perform quite well. This is commonly known as the exploration exploitation trade-off. There's deep research into this area. EasyBot has a little bit of secret sauce. Suffice it to say it's similar to how OpenAI makes their words more or less spicy.